Seizure prediction and blockage of seizures is a top research priority in epilepsy, a disease affecting about 1% of the US population. Due to the extreme complexity of brain dynamics, its analysis requires sophisticated methodologies and algorithms from applied mathematics, physics, and engineering. Much emphasis has been placed on techniques from nonlinear dynamics, including application of synchronization and dynamical invariants techniques. In this proposal, we will perform a systematic assessment of these techniques for the task of seizure detection and prediction on long time series. First, we look for optimal synchronization measures, examining phase synchronization, generalized synchronization, and unstable periodic orbit analysis, and assessing each for its sensitivity and specificity for prediction or detection of epileptic seizures. We will test these strategies as well Lyapunov exponents on multi-day, continuous ECoG recordings from several patients, with each recording containing several seizures.